A Decomposition-based State Space Model for Multivariate Time-Series Forecasting

TL;DR

DecompSSM is an end-to-end decomposition framework using parallel deep state space models to improve multivariate time-series forecasting by capturing trend, seasonal, and residual components more effectively.

Contribution

It introduces a novel decomposition-based deep state space model with adaptive scales and shared context refinement for multivariate time series forecasting.

Findings

Outperforms strong baselines on standard benchmarks

Effectively captures multiple temporal components

Demonstrates improved forecasting accuracy

Abstract

Multivariate time series (MTS) forecasting is crucial for decision-making in domains such as weather, energy, and finance. It remains challenging because real-world sequences intertwine slow trends, multi-rate seasonalities, and irregular residuals. Existing methods often rely on rigid, hand-crafted decompositions or generic end-to-end architectures that entangle components and underuse structure shared across variables. To address these limitations, we propose DecompSSM, an end-to-end decomposition framework using three parallel deep state space model branches to capture trend, seasonal, and residual components. The model features adaptive temporal scales via an input-dependent predictor, a refinement module for shared cross-variable context, and an auxiliary loss that enforces reconstruction and orthogonality. Across standard benchmarks (ECL, Weather, ETTm2, and PEMS04), DecompSSM outperformed strong baselines, indicating the effectiveness of combining component-wise deep state space models and global context refinement.